Higher Ethernet rates used to be introduced when industry bandwidth requirements drove the need for speed. Butwith Ethernet’s success, it soon became apparent that one new advance could satisfy the requirements of each Ethernet application space. This was clearly illustrated nearly 10 years ago when it was recognized that computing and networking were growing at different rates. This led to 40Gbps being selected as the next rate for servers beyond 10Gbps, while 100Gbps was selected as the next networking rate.
The industry, however, has a funny way of taking an Ethernet solution and applying it as it sees fit. Rather than use 40GbE as a server solution, the industry applied it in conjunction with 10GbE. This combination of 10GbE and 40GbE served to fuel the growth of hyper scale data centers, and while this has had a dramatic impact on the industry overall, it served to unchain Ethernet and lead to its next evolutionary step – introduction of rates targeted at applications, as opposed to force-fitting given Ethernet rates to given applications.
For example, 70 billion meters of deployed of Cat 5e and Cat 6 cabling, not capable of supporting the upgrade to 10GbE, led to the initiation of an Ethernet project to develop 2.5GbE and 5GbE BASE-T. These new specifications can be applied to the deployed cable infrastructure giving a needed speed-boost to campus and enterprise applications that are looking to deploy the next wireless standard, 802.11ac, that will drive the uplinks of wireless access points beyond today’s GbE uplinks.
So, while we saw 10GbE/40GbE data centers quickly being deployed, the industry recognized a couple of things. First, a higher density solution for 100GbE was necessary, and the easiest way to enable it was to reduce the width of the electrical interface from 10 lanes of 10Gbps to 4 lanes of 25Gbps. The industry also recognized the need for a higher speed of Ethernet beyond 100Gbps, which led to the industry effort to develop 400GbE.
The development of this higher speed Ethernet was different than the development of 40GbE and 100GbE, as new technologies were needed to develop a practical 400GbE solution. Figure 2 illustrates the options that can be considered during the development of new higher speed solutions, and how it was applied to the development of the target 400GbE specifications. More fibers running at 25Gbps was selected as the solution for operation over 100m of MMF. In the case of the solutions running over single mode fiber, operation at 50Gbps or greater was selected, and PAM4, a higher order modulation scheme, was chosen. And, while the 500m solution is running PAM4 at 100Gbps over 4 fibers, the 2km and 10km solutions added additional optical lambdas, and are running 8 lambdas at 50Gbps.
While the debates on the different technologies to enable 400GbE were underway, another dynamic was emerging. The development of 25Gbps signaling to support 100GbE, including operation over backplanes, copper twin-axial cables and multi-mode fiber, led to the recognition that 25GbE-based servers, using 100GbE for networking, could be used in the same fashion as 10GbE and 40GbE and drive the next generation of the hyper-scale data centers. The standard to make this leap will be completed shortly.
With the impending completion of this standard, and the rapid pace by which hyper scale data centers are looking to grow, it was recognized that, once again, a higher speed Ethernet was needed for the next generation of servers beyond 25GbE. And with the new 50Gbps signaling technologies being developed to support 400GbE, the choice is clear – 50GbE.
It raises another question, though: what would be the right networking solution Considering the success of 10GbE/40GbE, and the anticipated deployment of 25GbE/100GbE, it has been recognized that the optimal solution for servers is the highest serial signaling rate, while using a networking solution based on 4x this serial speed. Thus, the industry is now in the initial stages of starting the development of 50GbE and 200GbE.
Figure 3 illustrates what has emerged for the Ethernet industry. As discussed, Ethernet is no longer developing a single Ethernet rate to be applied everywhere and to everything. Instead, families of Ethernet rates, based on signaling rate technologies, are emerging. The base signaling rate and its quad: First generation is 10GbE and 40GbE; The next generation is 25GbE and 100GbE; And the latest generation of 50GbE and 200GbE.
As one looks at Figure 3, solutions based on 2x and 8x are logical and are emerging. And, as noted, 100Gb/s PAM4 optical signaling is in the development stage, thus a fourth generation data center architecture is clearly on the horizon. Furthermore, if one considers the development of an 8x solution, based on 100 Gbps, the next speed of Ethernet beyond 400Gbps is clearly 800Gbps.
Ethernet is continuing to evolve and the days of focusing on the development of a single speed are gone. Now, we see the industry focusing on multiple rates of Ethernet, intent on leveraging the investment in the next generation of signaling technologies to enable these multiple rates.
John D’Ambrosia chairs the Ethernet Alliance, a global consortium committed to the success and expansion of Ethernet technologies. He is also senior principal engineer at Huawei.